this affects longer documents which are each split into many chunks, correct? the number of documents we put into an index doesn't matter?

do people have experience of prior knowledge leaking into the response? not sure how to fix this as this is really unpredictable...
an example is:
idx_simple2 = GPTSimpleVectorIndex([])
idx_simple2.insert(Document("hello world"))
idx_simple2.query('tell me who is Stan Lee')
==> the results are obviously based on prior knowledge.

i'm trying to use GPTChromaIndex, but it does not seem to ingest the Document extra_info. is there a reason behind this?

i have multiple GPTSimpleVectorIndex and can find relevant documents when running index.query on each. But somehow if I query a combined list index, like GPTListIndex([index1, index2, index3]), it can't find relevant documents. I see an example of setting the summary for each index with set_text. Is this required for GPTListIndex to work? My indexes each contain a large number of documents so it does not seem reasonable to summarize all first.
https://gpt-index.readthedocs.io/en/latest/guides/use_cases.html#use-case-combining-information-across-multiple-indices